Online intelligent fault positioning method for vibration of transmission chain and tower drum of wind turbine generator

Abstract

The invention provides a wind turbine generator transmission chain and tower drum vibration online intelligent fault positioning method, and the method comprises the steps: collecting SCADA operation state data of a wind turbine generator, and obtaining vibration waveform data through vibration sensors disposed on a wind turbine generator transmission chain key mechanical part and a tower drum structure; performing multi-dimensional signal analysis on the vibration waveform data, and extracting a first fault feature set used for representing wind turbine generator transmission chain mechanical abnormality and a second vibration feature set used for representing tower drum structure dynamic abnormality; and combining SCADA operation state data, performing fusion analysis on the first fault feature set and the second vibration feature set, and performing fault positioning on vibration of the transmission chain and the tower drum of the wind turbine generator according to a preset diagnosis rule and a threshold value, thereby realizing online intelligent diagnosis and traceability of coupling vibration of the transmission chain and the tower drum, and improving the fault positioning accuracy of the coupling vibration of the transmission chain and the tower drum. And the accuracy, timeliness and operation and maintenance decision efficiency of fault positioning are effectively improved.

Description

Technical Field

[0001] This document relates to the field of wind turbine fault location technology, and in particular to an online intelligent fault location method for wind turbine drive train and tower vibration. Background Technology

[0002] The increasing size of wind turbine generators places higher demands on the reliability of key components, making online intelligent fault location particularly important. However, existing technologies for fault diagnosis of wind turbine drivetrains and towers still have certain limitations. For example, the solution disclosed in Chinese patent CN103411659B focuses on monitoring the structural state of blades and towers through vibration signals, but its method based on macroscopic parameter thresholds cannot pinpoint specific faulty components within the drivetrain. The integrated acoustic and vibration monitoring system disclosed in Chinese patent CN117007308A enhances the identification of early bearing defects by incorporating high-frequency acoustic signatures. However, this solution focuses on trend warnings for single components and does not address the problem of accurately locating fault sources in complex drivetrains.

[0003] When faced with complex faults involving the coupling of the drive train and the tower, existing methods struggle to determine the root cause and precise location of the fault through data fusion and correlation reasoning. Therefore, current technologies still fall short in terms of accurate fault tracing and intelligent location. There is an urgent need for an online fault location method capable of deeply integrating multi-source information and performing intelligent correlation reasoning to improve the accuracy of fault diagnosis and the efficiency of maintenance decisions. Summary of the Invention

[0004] This invention provides an online intelligent fault location method for vibration of wind turbine drive train and tower, aiming to solve the above-mentioned problems.

[0005] According to an embodiment of the present invention, an online intelligent fault location method for vibration of wind turbine drive train and tower is provided, comprising: S1. Collect SCADA operating status data of wind turbine units and obtain vibration waveform data through vibration sensors installed on key mechanical components of wind turbine transmission chain and tower structure. S2. Perform multi-dimensional signal analysis on the vibration waveform data to extract a first fault feature set for characterizing mechanical abnormalities in the wind turbine drive chain and a second vibration feature set for characterizing dynamic abnormalities in the tower structure. S3. Combining the SCADA operating status data, perform fusion analysis on the first fault feature set and the second vibration feature set, and locate the fault in the wind turbine transmission chain and tower vibration according to the preset diagnostic rules and thresholds. S4. Output a diagnostic report that includes fault location results, fault type, and severity.

[0006] The embodiments of the present invention have the following beneficial effects: By integrating SCADA operational data with multi-source vibration signals, collaborative analysis and intelligent fault location of drive train and tower vibrations were achieved, overcoming the limitations of traditional methods in accurately locating complex coupled faults. Through multi-dimensional feature extraction, adaptive normalization based on operating conditions, and an evolutionary diagnostic model based on deep learning and knowledge graphs, the accuracy of fault diagnosis and early warning capabilities were improved. The final output is a comprehensive diagnostic report including location results, type, severity, and confidence level, providing efficient and reliable decision support for preventative maintenance of wind turbine units, effectively reducing operation and maintenance costs and improving unit operational reliability. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart of the online intelligent fault location method for wind turbine drive train and tower vibration according to an embodiment of the present invention; Figure 2 This is a specific representation of a SCADA data field in an embodiment of the present invention; Figure 3 This is a schematic representation of vibration data measurement point information in an embodiment of the present invention. Detailed Implementation

[0009] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0010] Method Implementation Examples According to embodiments of the present invention, an online intelligent fault location method for vibration of wind turbine drive train and tower is provided. Figure 1 This is a flowchart of an online intelligent fault location method for wind turbine drive train and tower vibration according to an embodiment of the present invention. Figure 1As shown, the online intelligent fault location method for wind turbine drive train and tower vibration according to an embodiment of the present invention specifically includes: S1. Collect SCADA operating status data of wind turbine units and obtain vibration waveform data through vibration sensors installed on key mechanical components of wind turbine transmission chain and tower structure. The SCADA system collects macroscopic parameters of wind turbine operation. Data acquisition frequencies are divided into minute-level (10-minute) and second-level. The project uses a 10-minute data acquisition frequency. Based on SCADA data analysis, the unit's operating status is evaluated, such as... Figure 2 The diagram illustrates a specific SCADA data field representation according to an embodiment of the present invention. The SCADA operating status data includes, but is not limited to: wind speed, power, generator speed, various temperature data, grid current, and environmental parameters at the second or ten-minute level; among which, the temperature data covers key component measurement points such as main bearing temperature, gearbox oil temperature, and generator drive-end and non-drive-end bearing temperatures, used for subsequent temperature anomaly diagnosis and health assessment.

[0011] Vibration data reflects the microscopic health status of mechanical components. The project uses vibration data to identify mechanical defects such as pitting in gearboxes, wear on bearings, and misalignment of shafts, and locates faulty components through spectrum analysis. Figure 3 This is a schematic representation of vibration data measurement point information in an embodiment of the present invention.

[0012] The vibration data are acceleration waveform data collected by vibration sensors, with a sampling frequency of not less than 12800Hz to meet the requirements of high-frequency fault characteristic analysis. Vibration data measurement points include, but are not limited to: radial (horizontal and vertical) vibration points of the main bearing, input shaft, intermediate shaft, output shaft bearing housings, drive end and non-drive end bearing housings, bearing housings on both sides of the coupling, and triaxial vibration measurement points at the top, middle and bottom of the tower, used to comprehensively monitor the mechanical state of the transmission chain and the dynamic behavior of the tower structure.

[0013] In one specific implementation of this invention, the vibration sensors are configured and positioned as follows: On key mechanical components of the transmission chain, vibration sensors are arranged radially on the bearing seats of the main bearing (including horizontal and vertical directions), radially on the bearing seats of the input shaft, intermediate shaft, and output shaft of the gearbox, near the gear meshing point on the gearbox surface, radially on the bearing seats of the generator drive end and non-drive end, and on the bearing seats on both sides of the coupling. These sensors are used to monitor mechanical abnormalities such as rotational imbalance, shaft misalignment, bearing damage, and gear failure. On the tower structure, vibration sensors are respectively arranged at the top of the tower (near the nacelle floor), the middle of the tower (approximately halfway up the height), and the bottom of the tower (near the foundation ring). Preferably, triaxial vibration sensors are used to monitor the overall swaying, torsional vibration, and vibration transmission characteristics of the tower. The vibration sensors are preferably piezoelectric accelerometers with a frequency response range of not less than 0.5 Hz to 10 kHz, and are securely installed using magnetic bases or bolts to ensure the stability and accuracy of signal acquisition. The sampling frequency of the vibration data is not less than 12800 Hz to meet the requirements of high-frequency fault characteristic analysis.

[0014] S2. Perform multi-dimensional signal analysis on the vibration waveform data to extract a first fault feature set for characterizing mechanical abnormalities in the wind turbine drive chain and a second vibration feature set for characterizing dynamic abnormalities in the tower structure. The multi-dimensional signal analysis of the vibration waveform data includes time domain analysis, frequency domain analysis, envelope spectrum analysis, and trend analysis. The time-domain analysis includes: extracting time-domain statistical features from the vibration waveform data, wherein the time-domain statistical features include one or more combinations of mean, maximum, minimum, effective value, peak value, waveform, impulse, margin, skewness, and kurtosis.

[0015] The frequency domain analysis includes: converting the vibration waveform data to the frequency domain through fast Fourier transform to obtain a spectrum diagram; and calculating and labeling characteristic frequencies and their amplitudes related to the equipment rotation frequency and its harmonics, and the blade passing frequency based on the rotational speed information in the SCADA operating status data.

[0016] The envelope spectrum analysis includes: performing Hilbert transform on the vibration waveform data to extract the signal envelope, and performing spectral analysis on the envelope signal to generate an envelope spectrum; based on the geometric parameters and rotational speed information of the bearings in the transmission chain, calculating the bearing inner ring fault frequency, bearing outer ring fault frequency, rolling element fault frequency, and cage fault frequency, and marking and comparing the amplitudes in the envelope spectrum.

[0017] The trend analysis includes: periodically calculating preset time-domain, frequency-domain, or envelope spectrum characteristic values ​​for vibration waveform data acquired by one or more vibration sensors at continuous time points, and plotting the change curve of the characteristic values ​​with time as the horizontal axis to observe the long-term evolution trend of the mechanical state.

[0018] The extraction of the first fault feature set for characterizing mechanical anomalies in the wind turbine drivetrain and the second vibration feature set for characterizing dynamic anomalies in the tower structure specifically includes the following steps: For key mechanical components of the transmission chain, based on the results of time domain analysis, frequency domain analysis and envelope spectrum analysis, frequency features, amplitude features and modulation features related to rotational imbalance, shaft misalignment, component loosening, bearing failure and gear failure are extracted to form the first fault feature set; The first fault feature set may specifically include: time-domain indicators (such as RMS, kurtosis), frequency-domain parameters such as rotation frequency and its harmonics, blade passing frequency, amplitude or energy corresponding to the fault frequencies and harmonics of various bearing components, and amplitude at the bearing fault frequency in the envelope spectrum.

[0019] For the tower structure, based on the vibration sensor data arranged at different heights of the tower, the low-frequency vibration components, the energy changes within the preset range of the structure's natural frequency, and the vibration transmission phase relationship between multiple measuring points are analyzed to form the second vibration feature set.

[0020] The second vibration feature set may specifically include: the effective vibration value of each measuring point of the tower, the vibration energy near the natural frequency of the tower, the transfer function or coherence coefficient of the vibration at the top and bottom of the tower, and the modal characteristics reflecting the swing or torsion of the tower.

[0021] S3. Combining the SCADA operating status data, perform a fusion analysis on the first fault feature set and the second vibration feature set, and locate the faults in the wind turbine drivetrain and tower vibrations according to preset diagnostic rules and thresholds; the fusion analysis of the first fault feature set and the second vibration feature set specifically includes the following steps: Based on the SCADA operating status data, the first fault feature set and the second vibration feature set are subjected to operating condition normalization processing to eliminate the fluctuation of feature values ​​caused by different wind speed, power, and rotational speed. The operating condition normalization processing can be based on the wind speed, power, and rotational speed range in the SCADA data to map the vibration features to feature values ​​under standard operating conditions, or a machine learning model can be used to establish a normal state benchmark and calculate the deviation of the features from the benchmark.

[0022] Establish an association diagnosis rule base, which defines the combination logic and threshold conditions of multiple feature parameters in the first fault feature set and the second vibration feature set under a specific fault mode; The first fault feature set and the second vibration feature set, after normalization, are matched with the associated diagnostic rule base. Through multi-feature joint reasoning, the source of the fault is located.

[0023] Furthermore, the associated diagnostic rule base is constructed through the following steps: Using the first fault feature set, the second vibration feature set, and the corresponding SCADA operating status data after working condition normalization as input, a multimodal deep feature fusion network is constructed. The multimodal deep feature fusion network consists of parallel temporal convolutional network branches, graph neural network branches, and self-attention mechanism branches, which are used to extract the temporal deep features of vibration signals, the spatial topological correlation features between the transmission chain and the tower measuring points, and the long-term dependence and working condition coupling features of SCADA time series parameters. The multimodal feature fusion network maps the first fault feature set, the second vibration feature set, and SCADA operating status data to a shared deep feature space, generating a deep fault representation vector. The deep fault representation vector comprehensively represents the system-level health status of the drive train and tower coupling under specific operating conditions. Based on historical failure cases, expert diagnostic experience, and equipment physical models, an extensible failure knowledge graph is constructed. The failure knowledge graph uses failure modes as nodes and ternary relationships connecting failure nodes, typical patterns of deep feature vectors, and typical operating conditions that induce the failure as edges. Each edge is assigned a confidence weight and a dynamic adjustment factor. The deep fault representation vector generated online is matched and similarity calculated in real time with the fault knowledge graph. The matching process combines metric-based nearest neighbor search with graph neural network-based reasoning to locate the most likely fault mode and its propagation path. At the same time, the representation capability of the deep feature fusion network is continuously optimized through a contrastive learning strategy using online diagnostic results and subsequent operation and maintenance feedback. The weights and adjustment factors of the edges in the fault knowledge graph are dynamically updated using a reinforcement learning mechanism to achieve self-evolution of the diagnostic rule base.

[0024] The diagnostic rule base specifically includes diagnostic logic and thresholds for different fault types, for example: Rotational imbalance diagnosis: The amplitude of the 1X rotational frequency in the spectrum is significantly increased and is proportional to the square of the rotational speed. The corresponding 1X frequency component may also appear in the tower vibration.

[0025] Diagnosis of shaft misalignment: The amplitude of the 2X rotational frequency in the spectrum is increased, which may be accompanied by the 1X frequency component, and the vibration phase difference on both sides of the coupling is abnormal.

[0026] Bearing outer ring fault diagnosis: The bearing outer ring fault frequency (BPFO) and its harmonics in the envelope spectrum show obvious peaks, which may be accompanied by sidebands of the rotational speed frequency.

[0027] Gear wear / tooth breakage diagnosis: The gear meshing frequency and its sidebands appear in the spectrum, with the sideband interval being the rotational frequency of the shaft. The time-domain waveform may show periodic impacts.

[0028] Tower structure anomaly diagnosis: abnormal increase in vibration energy at the natural frequency of the tower, or significant change in the vibration transmission characteristics between measuring points at different heights of the tower, and abnormal correlation with wind speed and power in SCADA.

[0029] The first fault feature set and the second vibration feature set, after normalization, are matched with the associated diagnostic rule base. Through multi-feature joint reasoning, the source of the fault is located.

[0030] Furthermore, this embodiment of the invention further includes temperature anomaly diagnosis. Based on the temperature data in the SCADA operating status data, the multivariate state estimation technique (MSET) and the sequential probability ratio test (SPRT) method are used to monitor temperature anomalies and provide early warnings for faults in key parts of the wind turbine, as detailed below: Using at least one year of historical normal temperature data from wind turbines, an MSET baseline model was trained for each wind turbine and each key temperature measurement point.

[0031] The various types of temperature data collected in real time are input into the corresponding MSET model, the SPRT feature value score is calculated, the temperature status is judged as abnormal (normal, warning, alarm), and the abnormal time point, measurement point and SPRT score are recorded.

[0032] The SPRT scores of each temperature measuring point within a specified time period are calculated over time to generate a trend curve of temperature health status changes.

[0033] Furthermore, based on the SCADA operational status data, the MSET and SPRT methods are used to assess the health of each subsystem, specifically including: Identify the assessment subsystems: including at least the generator subsystem, engine room subsystem, power grid environment subsystem, and drivetrain subsystem.

[0034] Feature selection and weight allocation: Select representative SCADA feature parameters for each subsystem, and determine the weight of each feature in the subsystem using the CRITIC objective weighting method or expert experience method.

[0035] Health score calculation: For the feature data of each subsystem, the comprehensive SPRT deviation is calculated using its corresponding MSET benchmark model, and combined with the feature weights, the health score of the subsystem (e.g., 0-100 points) is calculated.

[0036] Overall health score: Based on the importance or preset weight of each subsystem, the health scores of the subsystems are weighted and averaged or merged to obtain the overall health score of the machine.

[0037] Furthermore, the obtained vibration fault location results, temperature anomaly diagnosis results, and subsystem and overall machine health assessment results are fused from multiple sources and jointly inferred. In the fault knowledge graph, the relationships between nodes and edges are expanded, associating typical temperature anomaly patterns and specific subsystem health decline patterns with specific vibration fault patterns and components. For example, an abnormally high temperature in the generator drive-end bearing, coupled with a decline in the generator subsystem health, should be associated with possible fault modes of that bearing in the vibration analysis (such as inner ring or rolling element failure). A comprehensive confidence assessment is performed on the diagnostic evidence from the three dimensions of vibration, temperature, and health. When evidence from multiple independent sources points to the same faulty component and type, the confidence score of the final diagnostic result is significantly improved. Based on the fused information, not only can the faulty component be located, but the root cause of the fault (such as long-term misalignment leading to bearing overheating and wear) and its development stage (early, middle, and late stages) can also be analyzed, and the comprehensive impact of the fault on the overall machine health status can be assessed.

[0038] S4. Output a diagnostic report that includes fault location results, fault type, and severity.

[0039] The method further includes: The confidence level of the fault location results is evaluated, and a confidence score corresponding to the fault location results is generated; When the confidence score is lower than the first preset threshold, the localization result is not output and is marked as an observation state. When the confidence score is not lower than the first preset threshold but lower than the second preset threshold, a diagnostic report containing the fault location result is output and its confidence score is attached, wherein the second preset threshold is higher than the first preset threshold.

[0040] The embodiments of the present invention have the following beneficial effects: By integrating SCADA operational data with multi-source vibration signals, collaborative analysis and intelligent fault location of drive train and tower vibrations were achieved, overcoming the limitations of traditional methods in accurately locating complex coupled faults. Through multi-dimensional feature extraction, adaptive normalization based on operating conditions, and an evolutionary diagnostic model based on deep learning and knowledge graphs, the accuracy of fault diagnosis and early warning capabilities were improved. The final output is a comprehensive diagnostic report including location results, type, severity, and confidence level, providing efficient and reliable decision support for preventative maintenance of wind turbine units, effectively reducing operation and maintenance costs and improving unit operational reliability.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An online intelligent fault location method for vibration of wind turbine drive train and tower, characterized in that... include: S1. Collect SCADA operating status data of wind turbine units and obtain vibration waveform data through vibration sensors installed on key mechanical components of wind turbine transmission chain and tower structure. S2. Perform multi-dimensional signal analysis on the vibration waveform data to extract a first fault feature set for characterizing mechanical abnormalities in the wind turbine drive chain and a second vibration feature set for characterizing dynamic abnormalities in the tower structure. S3. Combining the SCADA operating status data, perform fusion analysis on the first fault feature set and the second vibration feature set, and locate the fault in the wind turbine transmission chain and tower vibration according to the preset diagnostic rules and thresholds. S4. Output a diagnostic report that includes fault location results, fault type, and severity.

2. The method according to claim 1, characterized in that, The vibration waveform data is subjected to multi-dimensional signal analysis, including time domain analysis, frequency domain analysis, envelope spectrum analysis, and trend analysis.

3. The method according to claim 2, characterized in that, The time-domain analysis includes: extracting time-domain statistical features from the vibration waveform data, wherein the time-domain statistical features include one or more combinations of mean, maximum, minimum, effective value, peak value, waveform, impulse, margin, skewness, and kurtosis.

4. The method according to claim 2, characterized in that, The frequency domain analysis includes: converting the vibration waveform data to the frequency domain through fast Fourier transform to obtain a spectrum diagram; and calculating and labeling characteristic frequencies and their amplitudes related to the equipment rotation frequency and its harmonics, and the blade passing frequency based on the rotational speed information in the SCADA operating status data.

5. The method according to claim 2, characterized in that, The envelope spectrum analysis includes: performing Hilbert transform on the vibration waveform data to extract the signal envelope, and performing spectral analysis on the envelope signal to generate an envelope spectrum; based on the geometric parameters and rotational speed information of the bearings in the transmission chain, calculating the bearing inner ring fault frequency, bearing outer ring fault frequency, rolling element fault frequency, and cage fault frequency, and marking and comparing the amplitudes in the envelope spectrum.

6. The method according to claim 2, characterized in that, The trend analysis includes: periodically calculating preset time-domain, frequency-domain, or envelope spectrum characteristic values ​​for vibration waveform data acquired by one or more vibration sensors at continuous time points, and plotting the change curve of the characteristic values ​​with time as the horizontal axis to observe the long-term evolution trend of the mechanical state.

7. The method according to claim 1, characterized in that, The extraction of the first fault feature set for characterizing mechanical anomalies in the wind turbine drivetrain and the second vibration feature set for characterizing dynamic anomalies in the tower structure specifically includes the following steps: For key mechanical components of the transmission chain, based on the results of time domain analysis, frequency domain analysis and envelope spectrum analysis, frequency features, amplitude features and modulation features related to rotational imbalance, shaft misalignment, component loosening, bearing failure and gear failure are extracted to form the first fault feature set; For the tower structure, based on the vibration sensor data arranged at different heights of the tower, the low-frequency vibration components, the energy changes within the preset range of the structure's natural frequency, and the vibration transmission phase relationship between multiple measuring points are analyzed to form the second vibration feature set.

8. The method according to claim 1, characterized in that, The fusion analysis of the first fault feature set and the second vibration feature set specifically includes the following steps: Based on the SCADA operating status data, the first fault feature set and the second vibration feature set are normalized to eliminate the fluctuation of feature values ​​caused by different wind speeds, power, and rotational speeds; an association diagnosis rule base is established, which defines the combination logic and threshold conditions of multiple feature parameters in the first fault feature set and the second vibration feature set under a specific fault mode. The first fault feature set and the second vibration feature set, after normalization, are matched with the associated diagnostic rule base. Through multi-feature joint reasoning, the source of the fault is located.

9. The method according to claim 8, characterized in that, The associated diagnostic rule base is constructed through the following steps: A multimodal feature fusion module is constructed to fuse the first fault feature set, the second vibration feature set, and the SCADA operating status data, which have been processed by operating condition normalization, to generate a deep feature vector that comprehensively represents the system-level health status. Construct a fault knowledge graph module, where nodes represent fault modes and edges associate the fault mode with the typical patterns of the deep feature vectors that triggered the mode and the operating conditions. By matching and reasoning the online-generated deep feature vectors with the fault knowledge graph, fault patterns are located, and the feature fusion module and the knowledge graph module are continuously optimized using online diagnostic feedback.

10. The method according to claim 8, characterized in that, The method further includes: The confidence level of the fault location results is evaluated, and a confidence score corresponding to the fault location results is generated; When the confidence score is lower than the first preset threshold, the localization result is not output and is marked as an observation state. When the confidence score is not lower than the first preset threshold but lower than the second preset threshold, a diagnostic report containing the fault location result is output and its confidence score is attached, wherein the second preset threshold is higher than the first preset threshold.

Images